OpenAI's Reinforcement Fine-Tuning (RFT) is a technique that allows reasoning models like LLMs to adapt their thinking to new domains more effectively, and it can be applied to even more complex custom tasks. RFT uses reinforcement learning to update model weights, favoring generated outputs that receive higher grades, and it has been shown to work well with small datasets, reducing the data required by 1+ orders of magnitude compared to standard supervised fine-tuning. The technique is particularly useful for tasks where there is a clear "right" or "wrong" answer and outputs can be easily verified. RFT can be used as a stepping stone towards more optimized classical SFT models, especially for high-volume tasks, by first training an RFT model on a small dataset and then using it to machine-label additional examples before fine-tuning a simpler LLM using standard SFT. An open-source implementation of RFT is currently being developed, which aims to make the technique more accessible to researchers and practitioners.